Using automated paraphrasing tools: Examining the grammatical structure of generated paraphrases of scientific abstracts

Abstract

The proliferation of automated paraphrasing tools (APTs) raises questions about how generated paraphrases differ from the original texts. In this study, 50 abstracts published in Nature were compared with paraphrases generated by QuillBot, Jasper, and Copilot. Tactic and logico-semantic relations were analyzed using a modified version of the Hallidayan clause-complexing framework. The findings revealed that the APT that most closely matched Nature was QuillBot; no significant differences were found in any of the categories. Collectively, Jasper and Copilot leaned toward clausal complexity, using fewer paratactic extensions and more hypotactic elaborations. This study highlights general features of generated paraphrases that are not commonly addressed in the literature. Explicit paraphrasing instructions are recommended to avert any misuse of APTs.